Reproduce “Less is more: learning what to profile for GPU kernel optimization” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yihan Wang, Peiyu Liu, Runyu Chen, Wei Xu

AutoProfiler learns a deterministic tree to select GPU profiling measurements for each kernel before an LLM rewrites it. The tree routes kernels using earlier measurements, and the completed profile is passed to the rewriter once. On a controlled 48-kernel comparison, it scores 1.379× versus 1.314× for fixed feedback; on KernelBench, it reaches 1.704×, under the reported evaluation conditions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.